电商客服自动问答系统的商品意图识别

发布时间：2018-02-27 01:26

  本文关键词： 问答系统 电子商务 相似度 关键词 意图　出处：《五邑大学》2016年硕士论文　论文类型：学位论文

【摘要】：“中国制造2025”中提出要加快发展智能制造装备和产品,作为其中重要部分的服务机器人以及智能家居都倍受产业界追捧。在工业机器人之后,服务机器人也将得到政策支持,获得政策顶层设计规划。经过多年的发展我国的电子商务市场规模大,网络购物用户基数多,用户对商品的体验不再停留在产品质量,对服务质量的要求不断提高;另外,客服人员服务成本不断提升、流失率高以及招聘专业客服难度大等问题突出,电商客服机器人是人工智能的新形式,它将更高效地服务各商家和用户。客服人员以及电商客服机器人在和用户交流的过程中,关键环节在对用户意图的识别,只有准确发现用户的意图所在才能有效地为其服务,不断提高用户的满意进而获得忠诚的用户实现盈利。本文对电商客服自动问答系统(电商客服机器人)所属的问答系研究统进行了梳理,介绍了问答系统的发展以及系统的类型、结构等基础知识;同时,了解了问答系统涉及的关键词提取、词语相似度计算等技术,此外,深入理解了BP神经网络算法。在以上知识准备基础上,本文基于电商客服机器人系统,首先,在对用户语句处理上使用了中科院分词技术,并构建了化妆品领域所需的专业词典来提高分词准确性,词典涉及商品、美妆、护肤等多方面;根据网络购物语言特点,对用户沟通语料进行统计分析筛选建立了相应的停用词表;其次,选取语义、自身、位置三大块特征值信息利用BP神经网络模型对用户语句进行关键词提取。然后,本文构建了网络购物中用户比较关注的商品和服务意图网络,这一网络作为每一个用户的描述画像,为系统后期回答服务。最后,依据艾宾浩斯的人类遗忘规律,结合用户与客服机器人沟通的时间特点,基于遗忘曲线构建了“单阶段”和“多阶段”用户商品意图模型,实现用户的商品意图强度描述,通过提取的关键词利用Word2vec语义分析工具计算用户商品意图强度,对问答系统回答准确有一定改善,用户商品意图强度的获得还能给用户进行个性化推荐以及相关服务做指导。
[Abstract]:Service robots and smart homes, which are an important part of the "made in China 2025" initiative to speed up the development of intelligent manufacturing equipment and products, are popular in industry. After industrial robots, service robots will also receive policy support. After many years of development, the e-commerce market in China has a large scale, a large base of online shopping users, and the user's experience of goods no longer stays in the product quality, and the demand for service quality is constantly improved. The cost of customer service is constantly rising, the loss rate is high, and the difficulty of recruiting professional customer service is very serious. The e-business customer service robot is a new form of artificial intelligence. It will serve merchants and users more efficiently. In the process of communicating with customers, the key link of the customer service robot is the identification of the user's intention. Only by finding out exactly where the user's intention is, can it be served effectively. In this paper, the research system of the question and answer system, which belongs to the electronic customer service automatic question answering system (e-business customer service robot), is combed. This paper introduces the development of the question and answer system, the basic knowledge of the system, such as the type and structure of the system, and the key words extraction, word similarity calculation and other techniques involved in the question and answer system. Deeply understand BP neural network algorithm. Based on the above knowledge preparation, this paper based on the e-commerce customer service robot system, first of all, the use of Chinese Academy of Sciences word segmentation technology in the processing of user statements, The specialized dictionary needed in cosmetics field is constructed to improve the accuracy of word segmentation. The dictionary involves commodities, makeup, skin care and so on, according to the language characteristics of online shopping, The corresponding stop word list is established by statistical analysis and screening of the user communication corpus. Secondly, three blocks of eigenvalue information, namely, semantic, self and position, are selected to extract the key words of user statements by BP neural network model. This paper constructs a network of goods and service intentions that users pay more attention to in online shopping. This network, as a descriptive portrait of each user, serves for the later period of the system. Finally, according to the law of human forgetting, According to the time characteristics of the communication between the user and the customer service robot, a "single-stage" and "multi-stage" user's merchandise intention model is constructed based on the forgetting curve to describe the intensity of the user's commodity intention. The extracted keywords use Word2vec semantic analysis tools to calculate the strength of the user's commodity intention and improve the accuracy of the answer of the question and answer system. The acquisition of the intensity of the user's commodity intention can also provide users with personalized recommendation and relevant services to guide them.
【学位授予单位】：五邑大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP242;TP391.1
，

本文编号：1540545